Agentic AI is a term that describes systems configured to act autonomously based on dynamic reasoning, with little or no human intervention. The use of agent-based AI tools is becoming increasingly popular in the world of online shopping and payments, but the way agent-based AI systems work is at odds with existing payments rules, as Pinsent Masons’ David Tilbury highlighted earlier this month when exploring what new rules for agent-based AI in payments could look like.
This fact has now been acknowledged by the Treasury, which has said that existing payment service rules “fail to fully facilitate the use of agent-based AI.” He wants the UK to become a world leader in agency payments and has opened a consultation on modernizing the rules (43 pages / 453KB PDF), giving the industry a chance to have its say on how the framework should be adapted.
Pinsent Masons’ David Heffron, an expert in financial services regulation, said: “The Treasury consultation marks a decisive shift in the regulation of payment services in the UK.”
“Rather than simply updating existing rules, the government is proposing a more flexible model, with detailed requirements increasingly contained in the FCA Handbook, while the regulatory framework focuses on core principles, perimeter issues and consumer protection. Along with this structural reform, the consultation aims to forward-looking regulation of stablecoin payments, tokenized deposits, AI-powered agent payments and the next stage of open banking, signaling the UK’s clear ambition to become a global leader in payments innovation,” he said.
Regarding agent-based AI specifically, the Treasury said the UK has a “major opportunity to lead the development of agent payments globally.” It says the UK’s regulatory framework for payment services and e-money “must be designed to facilitate and support these innovations, while managing risks for consumers and businesses.”
In its separate plan for implementing artificial intelligence in financial services, published on Tuesday, the Treasury said establishing a “trust framework” to support the agency payments protocol was a “high” priority. It says agency payments are “a short-term practical analogue to a broader class of new autonomous financial systems.” As a result, it says creating the right framework to include agency payments could pave the way for “wider and more complex agency applications” in UK financial services.
Work to develop this new framework, according to the Treasury, should focus on three areas: defining clear legal frameworks and dispute resolution mechanisms to clearly allocate responsibilities when autonomous agents transact; creating standardized identification and verification systems specifically designed for AI and autonomous software agents; and creating interoperable technical standards that enable secure, seamless, and reliable machine-to-machine authentication. It says industry should lead the development of these new standards, but the government is open to legislating where necessary “to provide clarity and ensure safe implementation”.
Malcolm Dowden of Pinsent Masons, an expert in AI law, said: “The key issue in relation to payments made by agent AI is determining when a payment made by agent AI exceeds the authority given by the payer, whether due to hallucinations or purposeful behavior of the agent AI. The focus will therefore be on the interaction of the three provisions of the UK Payment Services Regulations 2017 – rules that establish consent or authority; rules that require the payer to act in accordance with the terms of the issuer.” and terms and conditions that allocate responsibility for unauthorized payments.”
“The key problem with agent-based AI is that it first needs to determine whether the system has exceeded the authority given to it by the payer before the issue of responsibility can be addressed,” he said.
In a speech at the Mansion House on Tuesday, Bank of England Governor Andrew Bailey said artificial intelligence was likely to boost Britain’s economic growth, but warned it would also “create legal problems”. He cited agent-based AI as an example and said there is a need to “proactively and positively” address the issues its use poses.
Bailey said: “Typically, the law makes the principal liable for the agent while the agent acts within the scope of the authority given to him by the principal. If the agent exceeds that authority or acts improperly, he becomes liable. Traditionally, this has depended on whether the agent has legal personality, i.e. whether the agent is a body corporate or a company, etc. But when the agent has no legal personality, does that mean the principal is always liable? If so, principals may be will have to constrain agents in ways that run counter to the direction of innovation in the AI Model.”
“In other words, is agent AI in this context more of a tool of the principal than strictly an agent, where the principal is ultimately responsible for everything done and not done? Does this apply when real money is spent in final and binding ways, as and when agent AI is allowed to hold funds? This removes questions about the boundaries of authority, but should force principals to think about how their control systems can work effectively with this technology. We need to think about these things now: and I use this example to illustrate why we must take a positive and proactive approach to solving problems that are critical to enabling effective innovation,” he said.
In her speech at the Mansion House dinner, UK Chancellor Rachel Reeves called AI “the defining technology of our generation” and noted its importance to the UK’s economy and national security, highlighting the leading role of the UK financial services sector in the adoption of AI, including in payments.
Last week the Financial Conduct Authority (FCA) published the findings of a study into artificial intelligence and the future of retail financial services led by FCA chief executive Sheldon Mills. Mills predicted that agent-based AI will change the way people access and use financial services, as well as the way the financial services industry operates. Among his recommendations, Mills called on the FCA to review the “regulatory perimeter”, identify potential regulatory gaps that could arise from the technological shift it anticipates, and adopt an “agency supervision model” that would represent a significant change in the way AI risks are addressed and the industry is regulated.